Security audit

bursar

Security checks for vulnerabilities and agentic risk

Overview

The skill coherently describes owner-limited payments and stock trading, with no hidden execution or unrelated data access.

Use a trusted Bursar MCP server and review the mandate's budget, providers, and permitted actions. The workflow can spend money, trade stocks, and collect due endowment draws. Job input goes to the selected provider, so exclude secrets and unnecessary personal information. The bundled instructions do not independently prove the server's enforcement claims.

A.I.G

Vulnerability Patterns

Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns

Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (2)

The payment workflow sends job input to an external provider without a privacy warning.

Medium

Category: Output Handling
Confidence: 88% confidence
Finding: The workflow instructs the agent to call mandate_pay_provider with an input object containing job content, such as a user prompt. Although it explains the movement of funds, it does not disclose that job content is shared with the external provider or warn against including sensitive information.
Content: No source excerpt is available for this finding.

The manifest describes reading an endowment, but L53–L54 additionally expose `endowment_draw` to collect a due draw.

Medium

Category: Excessive Agency
Confidence: 92% confidence
Finding: The manifest at L03 describes the endowment capability as read-only ('read your endowment'). Collecting a draw is a state-changing financial operation, not a status query, so the documented tool capability exceeds that stated scope.
Content: No source excerpt is available for this finding.